Election Typology: Theoretical Foundation and Empirical Measurement | EZstates

A Quantitative Framework for Presidential Election Classification: Theoretical Foundations and Empirical Measurement

EZstates Elections Research  ·  Working Paper  ·  2025
Data coverage: 2004–2024 U.S. presidential elections  ·  Groups: 135 demographic categories  ·  States: 50
Abstract
This paper presents a quantitative framework for classifying U.S. presidential elections using the typological tradition established by V.O. Key, Jr. (1955) and formalized by Campbell, Converse, Miller, and Stokes (1966). Prior applications of this typology have relied on aggregate vote returns and qualitative historical judgment. We introduce a scalar measure — the Contribution-Power-weighted mean absolute change in Partisanship Index (CP-weighted mean |ΔP|) — computed across 135 demographic groups observed in exit polls across all 50 states and six presidential elections (2004–2024). This yields 9,787 group-state-year observations and 3,794 cross-election comparison pairs. We classify each election cycle as Maintaining, Deviating, Converting, or Realigning based on empirically derived thresholds, and we separately operationalize the Switchers and Standpatters framework from Key (1966) at the group level. To our knowledge this is the first systematic, multi-group, multi-state quantification of the V.O. Key / Campbell et al. election typology using individual-level exit poll data.
Section 1

Theoretical Background

1.1 The Critical Election Concept

The systematic study of electoral change as a typological problem begins with V.O. Key, Jr.'s landmark article "A Theory of Critical Elections" (1955). Key observed that certain elections produce "a more or less sharp and durable reshuffling in the pattern of group support underlying the political parties" — a realignment that persists into subsequent electoral cycles. His empirical illustration using New England town-level returns from 1896 and 1928 established the precedent of tracking group-level partisan change across elections rather than treating each election as an isolated event.

Key subsequently distinguished between critical realignment, which occurs suddenly within a single electoral cycle, and secular realignment, a gradual drift in coalition support accumulating over multiple elections (Key, 1959). Both concepts share the core insight that electoral change is structured — it reflects durable shifts in how social groups relate to the party system — and that the magnitude and persistence of those shifts are the meaningful quantities.

1.2 The Four-Category Typology

The formal four-category classification of elections was developed by Angus Campbell in the chapter "A Classification of the Presidential Elections", published in the volume Elections and the Political Order by Campbell, Converse, Miller, and Stokes (1966). Building on the party identification framework introduced in The American Voter (Campbell et al., 1960) and the concept of the "normal vote" as a baseline, Campbell identified four types:1

Type Defining characteristics Persistence
Maintaining The majority party wins; group coalitions remain stable relative to prior election; normal partisan forces dominate Coalitions unchanged
Deviating The majority party loses due to short-term forces (candidate personality, salient issue); the underlying distribution of party identification is unchanged Returns to normal in subsequent elections
Converting The majority party wins, but through a meaningfully altered coalition — groups shift their relative support levels, producing a new alignment configuration that persists Persists; coalition composition changes structurally
Realigning Large-scale, durable reordering of group coalitions; typically involves substantial new voter mobilization and cross-party defection; the majority party relationship may change Persists for a generation; defines the subsequent political era

1.3 Switchers and Standpatters

A complementary micro-level framework appears in Key's posthumously published The Responsible Electorate (1966). Analyzing sequences of survey responses across consecutive elections, Key classified individual voters as Standpatters (who voted the same party in both elections), Switchers (who crossed party lines), and new or non-voters. His central argument — that switchers disproportionately determine election outcomes while standpatters maintain coalition continuity — provides the theoretical grounding for our cross-election group classification.

Walter Dean Burnham's Critical Elections and the Mainsprings of American Politics (1970) further elaborated the realignment framework, arguing that American electoral history is punctuated by periodic critical realignments separated by long periods of electoral stability. Burnham's work made realignment theory the dominant paradigm in electoral studies through the 1980s and remains an essential reference.

Section 2

Limitations of Prior Approaches

Despite the theoretical richness of the V.O. Key and Campbell et al. frameworks, empirical applications have faced two persistent limitations.

Aggregate-level measurement. Key's original analysis used town- and county-level vote returns. Campbell et al. relied on national survey distributions of party identification and vote choice aggregated to the national level. Neither approach can directly observe how specific demographic groups changed their partisan behavior across elections in specific states — the unit of analysis most relevant to electoral strategy and coalition theory.

Qualitative classification. The assignment of historical elections to the four types has proceeded largely through expert judgment applied to narrative evidence. Different scholars classify the same elections differently. There is no agreed-upon scalar measure that would allow a principled, reproducible determination of whether an election crosses the threshold from Deviating to Converting, or from Converting to Realigning.

Our framework addresses both limitations directly.

Section 3

Data and Measurement

3.1 Data Source

The empirical foundation of this framework is U.S. presidential election exit poll data, covering six elections: 2004, 2008, 2012, 2016, 2020, and 2024. Exit polls report, for each of 135 demographic groups, the share of the state electorate that group constitutes (group share) and the percentage of that group voting for the Democratic and Republican candidates (group vote shares). Data are structured at the group × state × year level.

135
Unique demographic groups
50
States
6
Election years
9,787
Group-state-year observations
3,794
Cross-election comparison pairs

Not all groups appear in all states in all years; exit poll coverage varies by sample size and state reporting practices. Observed coverage is 9,787 of a theoretical maximum of 40,500 (135 × 50 × 6), reflecting the structured sparsity inherent in exit poll data.

3.2 Partisanship Index (P)

For each group g, state s, and year t, we define the Partisanship Index:

Definition 1 — Partisanship Index
P(g, s, t) = (D(g,s,t) − R(g,s,t)) / 100

where D = % of group g voting Democratic, R = % voting Republican
Range: −1 (unanimously Republican) to +1 (unanimously Democratic)
When D + R < 85% (significant third-party share), a two-party variant P₂ = (D − R)/(D + R) is used.

3.3 Electoral Power Indices

A change in group partisanship matters for election outcomes in proportion to the group's size in the electorate. We decompose each group's electoral weight into two components:

Definition 2 — Contribution Power and Swing Power
CP(g,s,t) = (group_share / 100) × |P(g,s,t)|
SP(g,s,t) = (group_share / 100) × (1 − |P(g,s,t)|)

CP + SP = group_share / 100 (they partition total electoral weight)
CP measures the group's contribution of reliably partisan votes; SP measures its contestable votes. A large, strongly partisan group has high CP and low SP.

3.4 Cross-Election Change and the Classification Scalar

For each group observed in two consecutive exit-poll elections (t−1, t), we compute the absolute change in partisanship:

Definition 3 — Absolute Partisanship Change
|ΔP|(g, s, t) = |P(g, s, t) − P(g, s, t−1)|

We aggregate this to an election-level scalar using CP as the weighting factor, so that large, reliably partisan groups — which carry the most electoral weight — are weighted most heavily in the classification:

Definition 4 — CP-Weighted Mean |ΔP| (the classification scalar)
W(t) = Σ [CP(g,s,t) × |ΔP|(g,s,t)] / Σ CP(g,s,t)

where the sums run over all (g, s) pairs with valid observations in both years t and t−1
W(t) is the primary scalar used to classify election t relative to its predecessor. It ranges from 0 (no change) upward; values near 0.1 represent substantial realignment in historical context.

3.5 Group-Level Switchers and Standpatters

For each group-state pair observed in consecutive elections, we classify the cross-election behavior following Key (1966), adapted to the group (rather than individual voter) level:

Definition 5 — Group-Level V.O. Key Classification
dir(g,s,t) = sign(P(g,s,t)): +1 if Democratic-leaning, −1 if Republican-leaning

comb(g,s) = dir(g,s,t) + dir(g,s,t−1) ∈ {−2, −1, 0, +1, +2}

comb = +2 → Democratic Standpatters
comb = +1 → Weak Democratic (leaned Dem both elections; weakened)
comb = 0 → Switchers (crossed the partisan midpoint)
comb = −1 → Weak Republican
comb = −2 → Republican Standpatters
A group is a Switcher if and only if its partisan lean changed sign between the two elections — i.e., it was Democratic-leaning in one and Republican-leaning in the other.

3.6 Decisive Switches

A group's cross-election shift is classified as decisive if the net change in Democratic group votes exceeded the statewide winning margin:

Definition 6 — Decisive Switch Criterion
Let Δ(g,s,t) = Dem_group_votes(g,s,t) − Dem_group_votes(g,s,t−1)
Let M(s,t) = total Dem candidate votes(s,t) − total Rep candidate votes(s,t)

decisive(g,s,t) = TRUE iff [M > 0 AND Δ > 0 AND Δ > M]
OR [M < 0 AND Δ < 0 AND Δ < M]
The first condition covers a Dem-won state where the group's Dem vote growth exceeded the margin; the second covers a Rep-won state where the group's Dem vote decline exceeded the Republican margin. Both conditions identify groups whose shift was individually sufficient to account for the outcome.
Section 4

Election Classification Thresholds

We classify each election cycle using empirically observed values of W(t) across the six available election pairs. The classification boundaries are anchored to the distribution of W(t) values and to theoretical expectations about what magnitudes constitute meaningful versus transformative change:

Type Threshold (W) Interpretation
Maintaining W < 0.05 Group coalitions essentially stable; changes within normal sampling and short-term variation
Deviating 0.05 ≤ W < 0.08 Meaningful group-level shifts; consistent with short-term candidate or issue effects without structural realignment
Converting 0.08 ≤ W < 0.12 Substantial structural shifts across multiple groups and states; coalition composition changes that are likely to persist
Realigning W ≥ 0.12 Profound, broad-based transformation; large magnitude shifts across many groups and states; consistent with generation-level coalition change
Important Note on Thresholds
These thresholds are empirically derived from the distribution of W(t) values observed in 2004–2024 exit poll data, not derived from prior theory (which offered no scalar thresholds). With only five election pairs currently observed, the classification boundaries should be treated as provisional. As coverage extends to earlier elections or as additional exit-poll cycles are added, the thresholds should be revisited. The Deviating/Converting distinction in particular — which Campbell et al. (1966) grounded in whether shifts persisted into subsequent cycles — cannot be fully verified until future election data are available.
Section 5

Empirical Results: 2004–2024

Election Pair W(t) — Mean |ΔP| Classification Switcher groups Decisive switches ★ Interpretation
2004→2008 0.154 Realigning 65 Broadest coalition expansion in the exit-poll era; Obama mobilized new voters across racial, age, and educational groups
2008→2012 0.074 Deviating 21 Coalition consolidation; Obama won via high mobilization of the 2008 coalition rather than structural expansion
2012→2016 0.092 Converting 56 Trump realignment: working-class non-college white voters shift Republican; suburban college-educated voters shift Democratic
2016→2020 0.100 Converting 120 Highest decisive switch count in the series; suburban realignment deepened, record turnout amplified group-level shifts
2020→2024 0.060 Deviating 18 Broad but modest Republican gains across age, education, and Hispanic groups; whether these persist or revert determines the long-run classification

The results exhibit strong face validity. The 2004→2008 transition — the largest coalition shift in the data — correctly registers as Realigning, consistent with the near-consensus view of the Obama election in the literature. The 2008→2012 cycle registers as Deviating, consistent with consolidation rather than new realignment. The 2012→2016 and 2016→2020 cycles, both Converting, capture the symmetric nature of the Trump-era reshuffling: neither election by itself constitutes full realignment, but together they describe a structural transformation of the party coalitions. The 2020→2024 cycle correctly registers as the most modest shift of the series — significant in political terms, but falling below the Converting threshold.

Section 6

Our Contribution

The typological framework of Key (1955, 1966) and Campbell et al. (1966) has been applied descriptively for seventy years without a systematic, reproducible scalar operationalization. This work introduces three methodological advances:

1. Group-level, multi-state measurement. Rather than treating the electorate as a single national aggregate, we track 135 demographic groups independently across all 50 states. This captures geographic and demographic variation invisible to national-level analysis.

2. A continuous classification scalar. W(t), the CP-weighted mean absolute change in Partisanship Index, is a single reproducible number that places each election cycle on the Maintaining–Realigning continuum. Researchers using this framework will arrive at the same classification from the same data.

3. Group-level operationalization of Key's Switchers and Standpatters. We adapt Key's individual-voter typology to the demographic group level, identifying which groups switched partisan direction and whether those switches were individually decisive in determining state-level outcomes.

The framework is not a substitute for the theoretical distinctions drawn by Key and Campbell et al. In particular, the Deviating/Converting distinction — which in the original theory rests on whether shifts revert in subsequent elections — cannot be determined from a single election pair and requires longitudinal observation. Our classification should be understood as reflecting the initial magnitude of change; persistence is a separate empirical question that the growing dataset will progressively answer.

Section 7

Limitations and Caveats

Data Coverage
Exit polls do not cover all demographic groups in all states. Groups appearing in fewer than two election years within a state cannot contribute cross-election comparisons. Groups underrepresented in small states may have unreliable share and support estimates due to small sample cells within the exit poll.
Temporal Range
The framework currently covers six elections (2004–2024). Extending to pre-2004 elections — when exit poll standardization varied — would require careful harmonization of group definitions across survey instruments. Conclusions about long-run realignment dynamics are provisional.
Threshold Calibration
The classification thresholds (0.05, 0.08, 0.12) are empirically motivated but not theoretically derived. They produce face-valid classifications for 2004–2024 but may require recalibration as additional election cycles are added.
Decisive Switch Formula
The decisive switch criterion compares total Dem group vote growth against the statewide margin. This includes both the partisan shift effect and population change effects within the group. A cleaner measure would isolate only the vote change attributable to the support shift, removing demographic size effects. This refinement is planned for a future version.
References

References

  • Burnham, W.D. (1970). Critical Elections and the Mainsprings of American Politics. New York: W.W. Norton.
  • Campbell, A., Converse, P.E., Miller, W.E., & Stokes, D.E. (1960). The American Voter. New York: Wiley.
  • Campbell, A., Converse, P.E., Miller, W.E., & Stokes, D.E. (1966). Elections and the Political Order. New York: Wiley. [The four-category typology appears in Chapter 4: "A Classification of the Presidential Elections," by Angus Campbell.]
  • Key, V.O., Jr. (1955). "A Theory of Critical Elections." Journal of Politics, 17(1), 3–18.
  • Key, V.O., Jr. (1959). "Secular Realignment and the Party System." Journal of Politics, 21(2), 198–210.
  • Key, V.O., Jr. (1966). The Responsible Electorate: Rationality in Presidential Voting, 1936–1960. Cambridge: Belknap Press of Harvard University Press. [Introduces the Switchers/Standpatters framework.]
Notes
1 The precise formulation of all four categories including "Converting" as a distinct type should be verified against Campbell's Chapter 4 in the 1966 volume. Some secondary literature attributes only three categories (Maintaining, Deviating, Realigning) to Campbell et al. and treats "Converting" as a later addition or variant. Authors should confirm the original source text before citing the four-category typology as Campbell et al. (1966) in peer-reviewed work.
2 V.O. Key died in 1963. The Responsible Electorate (1966) was completed posthumously and edited by Milton Cummings, Jr.
3 The Standpatters/Switchers/Weak Democratic/Weak Republican five-category classification used in our cross-election analysis (Section 3.5) extends Key's original three-way typology (Standpatters, Switchers, New Voters) by distinguishing degrees of partisan stability among standpatters — an extension justified by the group-level nature of our data, in which a group that leans Democratic in both elections but less strongly so in the second year is meaningfully different from one that is uniformly dominant.